Mastra EUVD-2026-67804

Exploitation requires two conditions to be true simultaneously: (1) the attacker holds any valid authenticated session on the target Mastra instance (PR:L per the provided CVSS 4.0 vector - unauthenticated access is insufficient); and (2) the Mastra server is configured WITHOUT the mapUserToResourceId callback, which is the specific omission that disables thread ownership validation. Deployments that implement mapUserToResourceId correctly are not vulnerable regardless of version. Exploitation is bounded to confidentiality - read-only access to thread metadata and conversation history; no write, delete, or service-disruption capability is present. The attack is not limited to a specific platform or OS.

The CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:N/VA:N) indicates low-complexity network exploitation requiring only a low-privilege authenticated session, with high confidentiality impact and no integrity or availability consequence. No EPSS score is available in the provided data, limiting quantitative exploitation probability assessment. The absence of a KEV listing and confirmed POC suggests this is not yet exploited at scale; however, the VulnCheck disclosure makes the mechanism fully public, lowering the bar for future exploitation. Real-world risk is gated on deployment configuration: only Mastra instances omitting mapUserToResourceId are exposed, but VulnCheck's framing of this as a 'no-op' implies the omission may be common. Integrity and availability are unaffected, constraining the blast radius to data disclosure within a single Mastra deployment.

An authenticated user on a multi-tenant Mastra deployment issues a standard GET /api/memory/threads request and receives a full enumeration of threads belonging to all other users because the missing mapUserToResourceId configuration disables the ownership filter. The attacker then retrieves conversation histories and metadata from any returned thread ID, gaining access to potentially sensitive AI interaction data across the entire user base. No exploit code or special tooling is required - the vulnerability is exercisable with a standard HTTP client and any valid session token.